The Nuclear Option: Why AI's Real Bottleneck Is a 1979 Relic

Here is a specific data point that has been living rent-free in my head since I first ran the numbers: a single 100,000-GPU AI training cluster can draw peak power in the hundreds of megawatts. That is not a server room. That is a mid-sized city. For years, the narrative has been about chips, models, and scaling laws. But the forensic reality is that the silicon is irrelevant without a substation. The market is finally realizing this. Over the past 12 months, a peculiar thing happened on the way to the AI utopia: the smartest money in the world started buying power plants, not just GPU clouds. Constellation Energy (CEG), Talen Energy (TLN), Vistra (VST), and GE Vernova (GEV) have become the new picks-and-shovels plays. But the deeper story is not just about kilowatts; it is about a fundamental re-rating of a 1979 nuclear disaster site as the backbone of the next technological era. Signal in the noise: The most important infrastructure for AI might be a relic from the atomic age.